An Indian supercomputer by 2017. Umm…

This is a tricky question. And for background, here’s the tweet from IBN Live that caught my eye.

(If you didn’t read the IBN piece, this is the gist. India, rather Kapil Sibal, our present telecom minister, will have a state-of-the-art supercomputer, 61 times faster than current-leader Sequoia, built indigenously by 2017 at a cost of Rs. 4,700 crore across 5 years.)

Kapil Sibal

India already has many supercomputers: NAL’s Flosolver, C-DAC’s PARAM, DRDO’s PACE/ANURAG, BARC’s Anupam, IMS’s Kabru-Linux cluster and CRL’s Eka (both versions of PARAM), and ISRO’s Saga 220.

The most-powerful among them, PARAM (through its latest version), is ranked 58th in the world. It was designed and deployed by the Pune-based Centre for Development of Advanced Computing (C-DAC) and the Department of Electronics and Information Technology (DEITY – how apt) in 1991. Its first version, PARAM 8000, used 8,000 Inmos transputers (a microprocessor architecture built with parallel-processing in mind); subsequent versions include PARAM 10000, Padma, and the latest Yuva. Yuva came into operation in November 2008 and boasts a peak speed of 54 teraflops (1 teraflops = 1 trillion floating point operations per second; floating point is a data type that stores numbers as {significant digits * base^exponent}).

Interestingly, in July 2009, C-DAC had announced that a new version of PARAM was in the works and that it would be deployed in 2012 with a computing power of more than 1 petaflops (1 petalfops = 1,000 teraflops) at a cost of Rs. 500 crore. Where is it?

Then, in May, 2011, it was announced that India would spend Rs. 10,000 crore in building a 132.8-exaflops supercomputer by 2017. Does that make today’s announcement an effective reduction in budget as well as diminishing of ambitions? If so, then why? If not, then are we going to have two high-power supercomputers?!

Such high-power supercomputers that the proposed 2017-supercomputer will compete with usually find use in computational fluid dynamics simulations, weather forecasting, finite element analysis, seismic modelling, e-governance, telemedicine, and administering high-speed network activities. Obviously, these are tasks that operate with a lot of probabilities thrown into the simulation and calculation mix, and require hundreds of millions of operation per second to be solved within an “acceptable” chance of the answer being right. As a result, and because of the broad scale of these applications, such supercomputers are built only when the need for the answers is already present. They are not installed to create needs but only to satisfy them.

So, that said, why does India need such a high-power supercomputer? Deploying a supercomputer is no easy task, and deploying one that’s so far ahead of the field also involves an overhaul of the existing system and network architectures. What needs is the government creating that might require so much power? Will we be able to afford it?

In fact, I worry that Mr. Kapil Sibal has announced the decision to build such a device simply because India doesn’t feature in the list of top 10 countries that have high-power supercomputers. Because, beyond being able to predict weather patterns and further extend the country’s space-faring capabilities, what will the device be used for? Are there records that the ones already in place are being used effectively?